osx shadow server improvements: mdns, keyboard support, etc

[Antoine Martin]

The OSX shadow server needs similar fixes to the win32 one (#389), namely:

• keyboard
• faster screenscraping
• grab directly to gl? ​glgrab.c
• damage notifications:
  • ​CGRegisterScreenRefreshCallback
  • ​register for WindowServer CGScreenRefreshCallback CGScreenUpdateMoveCallback in C++
• maybe even export individual windows: ​Son of Grab / ​gdk_quartz_window_get_nswindow

[Antoine Martin]

Scheduling for 0.16

[Antoine Martin]

Re-scheduling.

[Antoine Martin]

Running from a non GUI terminal bails out with this ugly message, which should be turned into a more friendly init check:

xpra main error:
Traceback (most recent call last):
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/scripts/main.py", line 124, in main
    return run_mode(script_file, err, options, args, mode, defaults)
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/scripts/main.py", line 919, in run_mode
    return run_server(error_cb, options, mode, script_file, args)
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/scripts/server.py", line 927, in run_server
    app = ShadowServer()
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/platform/darwin/shadow_server.py", line 48, in __init__
    raise Exception("cannot grab test screenshot - maybe you need to run this command whilst logged in via the UI")
Exception: cannot grab test screenshot - maybe you need to run this command whilst logged in via the UI

[Antoine Martin]

Not sure if this is specific to OSX, but we should skip the stacktrace:

failed to create socket /var/tmp/502-Xpra/osx-2.local-0
Traceback (most recent call last):
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/scripts/server.py", line 357, in setup_local_sockets
    sock, cleanup_socket = create_unix_domain_socket(sockpath, mmap_group, socket_permissions)
  File "/Users/osx/Desktop/Xpra.app/Contents/Resources/lib/python/xpra/scripts/server.py", line 231, in create_unix_domain_socket
    listener.bind(sockpath)
  File "socket.pyc", line 228, in meth
error: [Errno 13] Permission denied
xpra initialization error:
 failed to create socket /var/tmp/502-Xpra/osx-2.local-0
closing tcp socket 0.0.0.0:10000

[Antoine Martin]

Some fixes:

• as of r11712 + r11714, the error from comment:4 looks like this:

  cannot grab pixels from the screen, make sure this command is launched from a GUI session
  

• r11714 also fixes sound forwarding from osx and win32 shadow servers (and therefore microphone forwarding too)
• r11711 removes the full stacktrace from the error in comment:5
• r11713 fixes the underlying error: the default xpra config was including a path specific to the machine it is built on! oops (part of the changes for #963)

---

New TODO: we should either ship the pybonjour mdns wrapper (as done with winswitch), or not bother trying avahi to avoid this message:

Warning: failed to load the mdns avahi publisher: No module named avahi
 either fix your installation or use the 'mdns=no' option

[Antoine Martin]

• r12246: the osx shadow server also has a systray
• r12231: publishes the shadow server via mdns using pybonjour

[Antoine Martin]

Important fix to backport (keypresses cause a crash!) in r.

The CGRegisterScreenRefreshCallback is deprecated in 10.8, instead we're supposed to use the ​Streaming the Contents of a Display API, which can send us frames in BGRA, YUV420, ..

Using ​Quartz API, CGRegisterScreenRefreshCallback should be able to replace the timer, and look like this (adapted from the pyobjc tests):

from Quartz.CoreGraphics import CGRegisterScreenRefreshCallback, CGRect
myInfo = object()
callcount = [0]
def callbackRefresh(count, rects, info):
    assert info is myInfo
    assert isinstance(rects, tuple)
    assert isinstance(count, (int, long))
    for i in rects:
        assert isinstance(i, CGRect)
        callcount[0] += 1
err = CGRegisterScreenRefreshCallback(callbackRefresh, myInfo)
assert err==0
...
CGUnregisterScreenRefreshCallback(callbackRefresh, myInfo)

Also useful:

• CGCursorIsVisible()
• CGCursorIsDrawnInFramebuffer
• CGWaitForScreenRefreshRects - crashes apparently
• ​pykeycode for converting keycodes to strings
• the toKeyCode function from this ​answer to "Generate keyboard events"
• ​How can I call CGEventKeyboardSetUnicodeString from python? for passing specific characters as key events?

The keymapping is not implemented yet..

[Antoine Martin]

• osx shadow server fixes: r12250 (crash!), r12254 (fix control-c hang)
• damage notifications implemented in r12252!

The printing stuff is blocked by #641: we need the installer to add the cups backend script to the system.

[Antoine Martin]

Note about selecting the audio device on OSX: ​osxaudiosrc/osaudiosink uses non-unique identifiers as device ID

[Antoine Martin]

A lot of progress in #641 / r12376:

• we have a PKG
• man page is installed
• cups backend (needs testing)
• /usr/bin wrappers

In order to support launching a remote shadow server via ssh (ie: xpra shadow ssh:OSXIP), we would need to run as an agent:

• ​Creating Launch Daemons and Agents
• ​Technical Note TN2083 - Daemons and Agents: Agents

Maybe the agent can just sit there, listening for events and we just tell it what to run (ie: xpra shadow "[email protected]"?) - which will then run from the correct context and be able to access the display server (avoiding the dreaded _RegisterApplication(), FAILED TO establish the default connection to the WindowServer, _CGSDefaultConnection() is NULL.): see #1169.

[Antoine Martin]

• r12390 switches to gstreamer 1.6 (see #1069)
• r12394 will make us use the osxaudiosrc ahead of autoaudiosrc
• r12395 + r12393 + r12396 + r12398: add very basic shadow keymap support (hardcoded) - will follow up in #1171
• r12397 disables printer forwarding with shadow servers (fails silently - why was I expecting anything more helpful from apple?) - will follow up in #1170
• r12391 fixes a hard crash in netifaces.gateways()
• r12392 should workaround NSCFData being passed to the network layer (but apparently does not.. will fix once I manage to reproduce reliably)
• the ability to start a shadow via ssh is in #1169
• keyboard improvements is now in #1172
• sound in #1205

@afarr: this is a FYI, feel free to close.

[alas]

Ok, thanks for info.

I actually gave it a shot, some luck with tcp, not so much with ssh... details of that failure in #1169.

This one looks like it's time it got closed.

[Antoine Martin]

You must be doing something wrong.
I've just re-tested against 10.6.x, 10.9.x and 10.10.x and it shadowed the existing display in all cases.

[Antoine Martin]

Follow up: #1321, #1322